According to some, Machine Learning is “the sexiest profession in the 21st century”. However, its media image, created based on science fiction, often deviates from the reality of working in this industry. Below, we dispel myths about ML and outline the key questions a budding ML developer should ask himself.
1. Do I need ML in my project?
ML is trendy. That doesn't mean you have to use it to accomplish your goals cheaply, quickly, and well. Of course, there are places where AI is the first choice. What's more, in addition to advantages such as great flexibility, it carries a lot of disadvantages and limitations. When deciding to use ML in a project, we must do it with our heads.
2. Do I understand the basics of mathematics well enough to understand ML?
Common computers, programming languages, and frameworks have incredibly simplified working with ML. We can clone and simply use someone else's repository, or code from a tutorial. With this approach, however, the result is black magic. We can't verify the correctness or evaluate the resultsólet alone consciously improve them and understand what the common problems are, such as a fading gradient, or techniques like regularization. The words ‘shit in, shit out’ are still true. You don't need to graduate in mathematics but pay attention to whether you understand why you are doing what you are doing.
3. Do I have the IT knowledge to simply code an AI project?
Creating AI is creating software. If you can't program fluently, use a version control system, get lost in an IDE and operating system, hold off! Packing ourselves with AI research problems when we are unfamiliar with the programming craft will end disastrously in the long run, and the level of fatigue and frustration will grow exponentially. The accumulating problems resulting from a lack of craftsmanship will discourage you from ML.
4. Are you aware of the division of labor in AI?
In practice, an AI project is about 80% work on data collection and preparation. Unless you work in a large and specialized team you will probably spend most of your time working with data. This requires knowledge of more than just programming. Again, mathematics comes into play, with which you be able to spot important relationships and problems in the data. You'll also need to develop a flair for the detective, the searcher who explores data poses questions and seeks answers.
5. Do you know what kind of knowledge you need?
In the past several years, with the development of Machine Learning, many free sources of knowledge have emerged. So the question often arises: is it worth it to choose a free course?"
Wandering around the web and searching for relevant, comprehensive, and reliable material can be time-consuming and inefficient. That time can be spent on family, trips, or sports. Education entrusted to professionals is certainly a good investment.
Summary
Machine Learning can be fascinating and challenging, but it can also be a generator of frustration and ineffective effort. We believe that knowing the scope of knowledge, the specifics.
